Client
Curated talent marketplace, United States
Engagement
AI evaluation layer, zero to MVP
Team
Jellie engineers embedded with the founding team
Timeline
Under 3 months to MVP
The starting point
The marketplace's pitch to clients was curation: every candidate on the platform is vetted, so hiring from it is faster and safer than the open market. The problem is that curation done by hand doesn't scale. Manual review was the bottleneck between the marketplace and its own growth.
The founding team came to us with an idea for AI interviewers. What they actually had was a business running on manual operations and a strong instinct that AI belonged at its core. Where, exactly, was still an open question.
What we built
The product itself
Before any build, we mapped their operations end to end: where hours went, what clients paid for, which manual steps carried the value. The answer changed the product. The leverage was in automated vetting, assessment, and matching, not in the interview. The roadmap, the story, and the automations below came out of that work.
Automated vetting pipeline
Candidates entering the marketplace move through an evaluation flow the moment they apply. What used to be a queue for manual review became a pipeline that runs on its own.
Skill assessment
Candidates are evaluated on the work they’d do, scored with reasoning the team can audit. Every candidate on the platform arrives assessed from day one.
Client-to-talent matching
When a client brings a role, the system scores it against the vetted pool and surfaces the strongest candidates with the reasoning attached.
Embedded, not outsourced
We worked inside the client's team from the first whiteboard: product definition, roadmap, and then the build. We brought the recruiting-AI judgment; they kept full ownership of direction and code.
The outcome
An MVP with AI evaluation at its core, shipped in under 3 months, and a different product than the one first imagined: vetting that scaled past manual review, assessments on every candidate, matching in from version one.
The engagement closed with a full handoff: infrastructure, accounts, and documentation transferred, with the client's team running the system independently.